A Risk-Based Assessment of Agricultural Water Scarcity Under Climate Change in a Semi-Arid and Snowmelt-Dominated River Basin

Transcription

1 Utah State University All Graduate Theses and Dissertations Graduate Studies 2016 A Risk-Based Assessment of Agricultural Water Scarcity Under Climate Change in a Semi-Arid and Snowmelt-Dominated River Basin Hossam Moursi Utah State University Follow this and additional works at: Recommended Citation Moursi, Hossam, "A Risk-Based Assessment of Agricultural Water Scarcity Under Climate Change in a Semi-Arid and Snowmelt- Dominated River Basin" (2016). All Graduate Theses and Dissertations. Paper This Thesis is brought to you for free and open access by the Graduate Studies at DigitalCommons@USU. It has been accepted for inclusion in All Graduate Theses and Dissertations by an authorized administrator of DigitalCommons@USU. For more information, please contact dylan.burns@usu.edu.

2 A RISK-BASED ASSESSMENT OF AGRICULTURAL WATER SCARCITY UNDER CLIMATE CHANGE IN A SEMI-ARID AND SNOWMELT- DOMINATED RIVER BASIN by Hossam Moursi A thesis submitted in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE in Civil and Environmental Engineering Approved: Jagath J. Kaluarachchi, Ph.D. Major Professor Mac McKee, Ph.D. Committee Member Thomas Lachmar, Ph.D. Committee Member Mark R. McLellan, Ph.D. Dean of the School of Graduate Studies UTAH STATE UNIVERSITY Logan, Utah 2016

3 ii Copyright Hossam Moursi All Rights Reserved

4 iii ABSTRACT A Risk-Based Assessment of Agricultural Water Scarcity Under Climate Change in a Semi-Arid and Snowmelt- Dominated River Basin by Hossam Moursi, Master of Science Utah State University, 2016 Major Professor: Dr. Jagath J. Kaluarachchi Department: Civil and Environmental Engineering Water scarcity is the major challenge that water managers face in semi-arid areas, especially in regions that depend on agriculture for rural livelihood. Climate change is one of the major stresses that is expected to exacerbate water scarcity problems in semiarid regions. In this study, a risk-based approach was used to assess the climate change impacts on the risk of agricultural water scarcity in semi-arid and snowmelt-dominated river basins that are dependent on agriculture. The Sevier River Basin, located in south central Utah, was used as the case study for this work. An agricultural water deficit index was proposed to represent the basin performance in terms of water supply and agricultural water demand. The basin s natural water supply was estimated using a semidistributed tank model. FAO AquaCrop model was used to estimate the crop water

5 iv requirements for major crops in the basin. The risk-based methodology begins using a vulnerability analysis to identify the system sensitivity to climate change. Sensitivity of system response to climatic variability was identified by establishing the climate response function, which is the relationship between basin agricultural water shortage and climate variables (i.e., precipitation and temperatures). The climate response function was then used to predict the basin agricultural water shortage in this century across four time slices using the projections of precipitation and temperature from downscaled and bias corrected GCMs outputs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) for RCP4.5 and RCP8.5 scenarios. The results of this study suggested that more natural water supply is expected in the Sevier River Basin due to the expected increase in precipitation during the future off seasons. However, projected temperature increases in the future may increase crop water requirements. It is also found that there is a high risk of unacceptable climate change impacts on agricultural water scarcity in the basin in the period under RCP4.5 and for under the RCP8.5 scenario, indicating climate change adaptation actions may be needed. (55 Pages)

6 v PUBLIC ABSTRACT A Risk-Based Assessment of Agricultural Water Scarcity Under Climate Change in a Semi-Arid and Snowmelt- Dominated River Basin Hossam Moursi Water shortage is a crucial concern in the semi-arid areas that have relatively low precipitation. Given the current situation in semi-arid areas where the water supply already does not meet the water demand, climate change is one of the important factors that is expected to exacerbate the water shortage problem and make the situation even worse in the future. There are many indicators that prove climate change is happening, like the increasing land temperature, sea level rise and decreasing snow cover. Therefore, studying climate change impacts is a vital issue, especially its impacts on water resources systems. In this research, we used a probabilistic approach to study climate change impacts on agricultural water in the Sevier River Basin, Utah. The used approach begins with identifying the basin sensitivity to climate change, and then using the projected temperature and precipitation from climate models to estimate the future agricultural water shortage. The results of this research indicated that the Sevier River Basin system is very sensitive to precipitation and temperature changes, and consequently, to climate change.

7 vi The basin s natural water supply is expected to increase as a result of future precipitation. In addition, the agricultural water demand is expected to increase due to rising temperature. It is indicated also that the basin will have a high risk probability of having agricultural water shortage in the period and

8 vii ACKNOWLEDGMENTS I would like to thank God for his grants that he bestowed on me and for giving me the ability and power to accomplish this research. I truly like to express my deepest gratitude and appreciation to my major advisor, Dr. Jagath J. Kaluarachchi, for advising my research and for his patience and support. No words of thanks can sum up the gratitude that I owe to Dr. Kaluarachchi for his guidance and scientific opinions which steered me in the right direction. I would also like to thank my committee members, Dr. Mac McKee and Dr. Thomas Lachmar, for their excellent reviews and valuable feedback. Many thanks to Dr. Daeha Kim for his assistance in better understanding the Sevier River Basin system and for his cooperation in establishing the algorithm of AquaCrop and Tank models. Additional thanks to USAID for funding my academic scholarship. I acknowledge the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP s Working Group on Coupled Modeling (WGCM) for their roles in making available the WCRP CMIP5 multimodel data set. I am very thankful to my great family in Egypt. My beloved mother, brothers and sisters, thank you so much for all you have done for me over the years. I could not have made it without your love and support. Numerous thanks to my deceased father for his great role in my life and his sacrifices for me and my family. Without the support of those all mentioned above, my thesis would not have been possible. Hossam Moursi

9 viii CONTENTS Page ABSTRACT... iii PUBLIC ABSTRACT... v ACKNOWLEDGMENTS... vii LIST OF TABLES... ix LIST OF FIGURES... x CHAPTER 1. INTRODUCTION STUDY AREA AND DATA Description of Study Area Climate and Soil Data METHODOLOGY Proposed Framework for Climate Change Impact Assessment Stochastic Generation of Climatic Observations Estimation of Annual Water Supply Estimation of Irrigation Water Requirements GCMs and Downscaling Methods RESULTS AND DISCUSSION Climate Response Function Climate Change Impacts Sensitivity Analysis SUMMARY AND CONCLUSIONS REFERENCES... 41

10 ix LIST OF TABLES Table Page 1. Details of SNOTEL and NOAA stations located in different sub-watersheds and farming regions in the Sevier River Basin Irrigated crop areas and soil properties for each farming region in the Sevier River Basin Information of 31 and 29 GCMs used in the analysis for RCP4.5 and RCP8.5 scenarios respectively, with NA indicating non-availability Percent change in future precipitation and temperature for RCP4.5 and RCP8.5 scenarios Probability of unacceptable threshold (It < 0) for percent changes of groundwater use from 2010 for the given time periods using RCP4.5 and RCP8.5 projections Probability of unacceptable threshold (It < 0) for different irrigation efficiencies using RCP4.5 and RCP8.5 projections

11 x LIST OF FIGURES Figure Page 1. Physical layout of the Sevier River Basin, Utah Flow chart illustrating the proposed methodology for climate risk assessment A graph illustrating the algorithm for calculating the agricultural water requirement in AquaCrop Projections of precipitation during the off-season (Po) for the RCP4.5 scenario Projections of precipitation during the growing season (Pc) for the RCP4.5 scenario Projections of maximum temperature during the growing season (Tcmax) for the RCP4.5 scenario Box plot of future agricultural water deficit index (It) predicted by GCM projections for RCP Box plot of future agricultural water deficit index (It) predicted by GCM projections for RCP Scatterplot of % change in annual mean precipitation and annual maximum temperature during the growing season, superimposed on the contour lines of It for RCP4.5 and RCP8.5 scenarios when Po is constant at mm Scatterplot of % change in annual mean precipitation during the off-season and maximum temperature during the growing season, superimposed on the contour lines of It for RCP4.5 and RCP8.5 scenarios when Pc is constant at mm Cumulative distribution function of It calculated from GCM projections for the RCP4.5 scenario Probability of acceptable and unacceptable water supply scenarios for the RCP4.5 scenario Probability of acceptable and unacceptable water supply scenarios for the RCP8.5 scenario

12 CHAPTER 1 1. INTRODUCTION Planning and management of water resources in semi-arid regions is a major challenge due to water scarcity. A semi-arid region is defined as a region that has an average annual precipitation of mm/year and where evaporation is always larger than precipitation. Water scarcity happens when the water demand from all sectors (e.g., agriculture, municipal and industrial, etc.) exceeds the available water supply. Approximately, 30% of the world land area is considered arid and semi-arid and 20% of the world s population lives in these areas (Sivakumar et al. 2005). In most parts of the world, present water demand is already well in excess of supply and many more areas are expected to experience water scarcity as the world population continues to rise (Gourbesville 2008). Currently, one third of the world s population lives under high water stress and it is expected to increase to two thirds by 2025 (FAOWATER 2015). Agricultural production plays a major role in ensuring food security. Most of the rural semi-arid areas are dependent on agricultural productivity, as 70% of the total global fresh water withdrawals are used for irrigation (Fischer et al. 2007). Therefore, the major concern for water managers of these regions is the efficient management of water use for agriculture. While semi-arid basins already face water scarcity at the present, there are a number of factors that elevate the water shortage problem and will make the situation worse in the future. Some of the factors that may cause serious water stress in semi-arid

13 2 basins are economic growth, rapidly increasing population, high agricultural water use, groundwater depletion, and climate change. Climate change is a top priority of these factors, which is expected to increase water scarcity and the frequency of extreme conditions, and is expected to have a serious impact on irrigated agriculture. Irrigated agriculture in arid and semi-arid watersheds is sensitive to climate change (IPCC 1995). A plenty of scientific evidences confirm that climate change is already happening and one such observation is the global warming observed during the past 100 year (IPCC 2007), as the global average surface temperature has increased by 0.74 C over the past 100 years ( ). Therefore, climate change impacts on water resources systems have to be adequately addressed for sustainable water management. Climate change impacts on water resource systems are commonly assessed by retrieving the climate information projections from general circulation models (GCMs). GCMs simulate the physical processes in ocean, land surface, and atmosphere to measure the impact of increasing concentrations of greenhouse gases on the global climate system. GCMs have significant uncertainty in simulating climate change because they do not capture the physical processes which occur at relatively fine resolution. Therefore, the potential hydrological impact of climate change plays a significant role in the uncertainties of determining the future water demand and availability (Middelkoop et al. 2001). Despite these uncertainties associated with determining the magnitude and timing of climate change impacts, water managers seek to identify these regional impacts on

14 3 their systems given the importance of climate change in water resources (Snover et al. 2003). There are two major approaches for evaluating climate change impacts on water resources systems. The first approach is the top-down approach which is the most commonly used approach. It begins by using the climate projections of precipitation and temperature from GCMs as inputs to a hydrologic model. Since these projections typically use low spatial (100s of km) and temporal (monthly) resolution grid cells, a downscaling approach should be used to convert these results from low resolution to relatively high resolution to match the resolution of the hydrologic model. The output of the hydrologic model is then used to drive a water resource model to assess the climate risk on water resources. This approach is popular and has been applied by many researchers, especially on evaluating the climate change impacts on irrigated agriculture (e.g., Doll 2002; Stöckle et al. 2010; Vano et al. 2010; Gondim et al. 2012; Mainuddin et al. 2014). Doll (2002) used this approach to present the first global analysis of climate change impact on irrigation requirements. Stöckle et al. (2010) studied the potential impact of climate change on the crop yield for four selected crops in eastern Washington. Vano et al. (2010) used a reservoir system model combined with a hydrologic model to study the impact of climate change on irrigated agriculture and water management in the Yakima River Basin, Washington. Gondim et al. (2012) studied the climate change impact on the irrigation water needs in the Jaguaribe River Basin in Brazil using the Hadley Centre Regional Climate Model. They used the PRECIS (Providing Regional Climates for Impacts

15 4 Studies) system to generate the climate projections using a dynamical downscaling method. Mainuddin et al. (2014) studied the climate change impact on the irrigation requirements for the main crops in Bangladesh using a soil water balance simulation model (SWB). The limitation of this approach is that it increases the uncertainty associated with evaluating the climate change impacts, because every climate model and a corresponding emission scenario gives different results. In addition, this approach typically uses a small selection of climate models and scenarios because of the computational burdens for downscaling the GCM results and simulating the hydrologic model for each climate model and scenario combination. As a result, this approach does not capture the full range of future climate conditions (García et al. 2015). Therefore, a probabilistic approach could be more appropriate for assessing the climate change impacts to fully analyze and capture the uncertainty associated with different GCMs and emission scenarios. An alternative approach is the bottom-up decision-scaling methodology described by Brown et al. (2012). It integrates a bottom-up stochastic analysis to assess the system vulnerability with top-down use of GCM projections to estimate the relative probability of a future climate hazard. This approach consists of three steps. The first step is the characterization of the system and its performance indicators (e.g., system reliability) and identification of the threshold of acceptable performance that does not need a remedial action and unacceptable system performance that requires a further action. The second step is using a stochastic analysis to establish the climate response function to identify the

16 5 system climate sensitivity. The climate response function is the relationship between the system performance indicator and the climate variables. The third step is using different datasets of GCM runs to estimate the future values of the system performance indicator. Then the relative probability of each climate state of the acceptable and unacceptable performance can be estimated. The advantage of this approach is that it can use a large number of GCM projections and scenarios without conducting the analysis for each projection dataset. It can capture the full possible range of GCMs outputs and reduces the associated uncertainty in assessing the climate change impact while minimizing the computational effort. This approach is innovative and Brown et al. (2012) applied it to a simple municipal surface water supply system that included a supply reservoir. Ghile et al. (2014) used the decision-scaling approach to assess the climate change impact on infrastructure investment in the Niger River Basin. Li et al. (2014) demonstrated the same approach in the analysis of climate change impacts on a large water supply system in Massachusetts, USA. An example of rural semi-arid basins that is dependent on agricultural economy is the Sevier River Basin in Utah. It is located in south central Utah in the southwestern United States, which is considered the most arid region in the US, and is expected to face more water scarcity and more intense and frequent droughts (Seager et al. 2007). In addition, the basin water supply is driven from snowmelt, as winter precipitation is stored in the snowpack and discharged in the water demand season. Approximately 70-80% of the total runoff in the mountainous regions of the western US is derived from snowmelt

17 6 (Doesken and Judson 1997). Since climate change is expected to decrease snowpack accumulation and cause earlier snowmelt runoff, reduced water availability can cause serious water shortages during the growing season in these snowmelt-dominated regions (Llewellyn and Vaddey 2013). In this research, we extended the methodology proposed by Brown et al. (2012) to assess the climate change impact on agricultural water scarcity in the Sevier River Basin. The Sevier River Basin, Utah is used as the test case for this demonstration. This thesis is structured into 5 chapters where Chapter 1 is the introduction. Chapter 2 contains a description of the Sevier River Basin study area and the data used in this research. Chapter 3 describes the research methodology and the simulation models used. Chapter 4 presents the research results and discussion. Chapter 5 provides the research summary and conclusions.

18 7 CHAPTER 2 2. STUDY AREA AND DATA 2.1 Description of Study Area The study area is the Sevier River Basin located in south-central Utah (Figure 1). It encompasses approximately an area of 27,389 km 2 (12.5% of Utah s area) with a relatively high evapotranspiration (ET) and low precipitation, and is characterized as snowmelt-driven streamflows due to high elevation. The mean annual precipitation ranges between 250 and 1000 mm along the elevation profile. Particularly, water from the Sevier River has been highly regulated for irrigating farm lands, developed along the main channel and its tributaries, mainly by three reservoirs (Piute, Otter Creek, and Sevier Bridge Reservoirs). Most of the water supply from the reservoirs is used for agricultural production for rural livelihood; thus, decreasing snowfall and early spring runoff are important challenges for efficient water management, especially under climate change. Major irrigated crops are pasture and grass hay (45%), alfalfa (44%), maize (6%), barley (4%), and wheat (1%). Productivity of crops is significantly dependent on water availability from the three reservoirs during the growing season, which is from April to September. There exist municipal and industrial water demands, but these demands are small in comparison to the dominant agricultural water demand. In the Sevier River Basin, streamflows mostly originate from the upper watersheds in higher elevations, but water is consumed mostly by farm lands in the lower elevations given the higher water right according to the prior appropriation doctrine of

19 8 the western USA. Hence, farm lands near Delta have well supplied water from storage in the reservoirs. Farm lands near Fillmore are outside the geologic boundaries of the Sevier River Basin, but are irrigated by surface water from the Sevier River transported via the Central Utah Canal, as well as local groundwater sources. For the climate change impact assessment, the upper basin above the Sevier Bridge Reservoir is divided into 23 watersheds for simulating streamflows that account for most of the natural water supply to the reservoirs. Runoff from watersheds below the Sevier Bridge Reservoir is neglected due to high ET losses. To estimate the agricultural water demands, the farm lands along the main channel and tributaries are represented by eight regions: Delta, Oak, Ephraim, Fillmore, Richfield, Angle, Circleville and Panguitch, as shown in Figure Climate and Soil Data We used daily precipitation, and maximum and minimum temperatures from 1994 to 2015 (22 years) at 12 Snow Telemetry (SNOTEL) stations operated by the U.S. Department of Agriculture (USDA) for simulating natural flows in the 23 watersheds (available at We additionally collected daily precipitation and temperatures with the same length from the National Climatic Data Center (NCDC; available at of the National Oceanic and Atmospheric Administration (NOAA) for estimating irrigation requirement using AquaCrop in the eight farming regions. The number of NOAA stations used in this study was six. In total, 18 climatic stations located across the Sevier River Basin were used for

20 Figure 1. Physical layout of the Sevier River Basin, Utah. 9

21 10 this study, and are listed along with their corresponding watersheds and farming regions in Table 1. However, for generating the 10,000 year stochastic simulations, we used the data form the total 31 stations located in the basin to preserve the spatial correlations between them. Due to the absence of wind speed and relative humidity data with the same length of precipitation and temperature datasets, we used a constant wind speed obtained from nearby USDA Soil Climate Analysis Network (SCAN) stations and minimum temperatures for estimating reference ET in the eight farming regions. The soil physical properties required for estimating irrigation water requirement using AquaCrop are saturated hydraulic conductivity (Ksat), saturated water content (Өsat), field capacity (Өfc), bulk density (BD), permanent wilting point (Өwp) and electrical conductivity (EC). These data were obtained from SSURGO (Soil Survey Geographic Database) and STATSGO2 (State Soil Geographic Database) provided by the USDA Natural Resources Conservation Service. For estimating cropping area and crop classification, we used the CropScape data of USDA given at m 2 spatial resolution. The area of the five major crops in the basin in each farming region and the soil properties of each farming region are shown in Table 2. It should be noted that we refer to the crop areas here as crops that are irrigated from surface water supply only. We estimated the crop areas that are irrigated from groundwater supply by multiplying the total crops area by the percent of groundwater use; and subtracted it from the total crop area.

22 Table 1. Details of SNOTEL and NOAA stations located in different sub-watersheds and farming regions in the Sevier River Basin. 11 Sub-watershed SNOTEL Station Farming Region NOAA Station Asay Creek Midway Valley Delta Delta Mammoth Creek Castle Valley Oak Delta Panguitch Creek CastleValley Ephraim Ephraim USFS Pass Creek Sevier River Long Valley JCT Fillmore Delta Bear Creek Sevier River Richfield Radio Big Flat Richfield KSVC City Creek Sevier River Big Flat Angle Angle Upper East Fork Sevier River Agua Canyon Circleville Circleville Middle East Fork Sevier River Widtsoe 3 Panguitch Panguitch Lower East Fork Sevier River Widtsoe 3 Upper Otter Creek Box Creek Lower Otter Creek Box Creek Beaver Creek Sevier River Kimberly Mine Clear Creek Kimberly Mine Cottonwood Creek Sevier River Box Creek Lost Creek Sevier River Farnsworth Lake Salina Creek Pickle KEG Willow Creek Pickle KEG Silver Creek Mammoth Cottonwood Upper San Pitch River Mammoth Cottonwood Middle San Pitch River Seeley Creek Lower San Pitch River Seeley Creek Twelvemile Creek Beaver Dams Sevier Bridge Beaver Dams

23 12 Table 2. Irrigated crop areas and soil properties for each farming region in the Sevier River Basin. Delta Ephraim Circleville Panguitch Angle Richfield Fillmore Oak Crop Area (acres) Barley Wheat Corn Alfalfa Pasture Soil properties K sat (mm/day) Ө fc Ө wp BD (gm/cm 3 ) Ө sat EC (ds/m)

24 13 CHAPTER 3 3. METHODOLOGY 3.1 Proposed Framework for Climate Change Impact Assessment For assessing climate change impacts on basin-wide agricultural water scarcity, we adopted the decision-scaling approach proposed by Brown et al. (2012). While a typical climate change impact assessment uses GCMs as inputs of system models (e.g., runoff or crop models), the decision-scaling approach evaluates system sensitivity to climatic variability first and provides the degree of climate change with respect to the system sensitivity. If climate is not sensitive to the system behavior or specific system performance outputs, then further action may not be needed. In the case of high climate sensitivity, then the analysis is continued to assess the actual relevant climate change impacts. The proposed approach consists of three steps, which are shown in Figure 2. The first step is to identify the system models and corresponding performance indicators together with one or multiple thresholds of acceptable performance. The system models used in this study are the semi-distributed Tank Model similar to the work Cooper et al. (2007) for estimating natural water supply, and FAO AquaCrop (Steduto 2009), and the approach of Masoner et al. (2003) for estimating irrigation water requirements. Agricultural water deficit is the major concern for water managers in the Sevier River Basin, since the largest portion of water is used for irrigation purposes. Therefore, an annual agricultural water deficit index (It) is defined as the system performance indicator. It is defined using annual water availability (St) and agricultural

25 14 Identify System Model and Performance Indicator Identify performance indicator as Agricultural Water Deficit Index I t = S t D t D t Identify risk threshold for water scarcity as: I t = 0 Assess Sensitivity of System Response to Climate Change Stochastic simulation of daily P and T for 10,000 years using historical data Estimate basin water availability (S t ) Estimate basin agricultural water demand (D t ) Develop climate response function I t =f (P, T) Estimate Probability of Future Climate Risk for Water Scarcity Project P and T under climate change using GCMs Estimate I t under projected P & T and identify GCMs failing threshold of I t 0 Estimate relative probability of climate risk producing water scarcity across all GCMs Figure 2. Flow chart illustrating the proposed methodology for climate risk assessment, where P and T are basin precipitation and temperature, respectively.

26 15 water demand (Dt): I t = S t D t D t (1) where, St is total annual volume of streamflow in year (t) estimated by the semidistributed Tank Model (Mm 3 ), and Dt is total volume of seasonal irrigation water required (Mm 3 ) for attainable crop yields simulated by AquaCrop and the approach of Masoner et al. (2003). Attainable crop yield is defined here as the yield that can be reached under good water management practices. More details of this definition will appear later. A higher It value indicates a lower agricultural water shortage and vice versa. When It is equal to or greater than zero, no agricultural water deficit exists (i.e. natural water supply meets the agricultural demand). Therefore, we defined the threshold of the acceptable basin performance as It = 0. The second step is to identify the sensitivity of system response to climatic variability by sampling a broad range of possible climate change scenarios using a stochastic simulation. To accommodate a broad sampling range of climatic conditions, we first extended the relatively short climatic data (i.e. 22-year precipitation and temperatures) to 10,000 years using multi-site stochastic models. St and Dt are simulated using stochastically generated 10,000-year precipitation and temperatures as inputs to the system models, and therefore It values for 10,000 years are computed from the system models. Using these It values and annual precipitation and temperature values, a relationship between It and climate conditions, temperature and precipitation, can be established. This relationship is typically called the climate response function.

27 16 In step 3, multiple GCMs are used as inputs to the climate response function to estimate the future probability distribution of agricultural water deficit index. The climate response function provides the ease of including multiple GCMs in the analysis. Once the climate response function is established, no additional simulation using the system models is necessary. Hence, agricultural water scarcity corresponding to a GCM can be quickly found using the climate response function. By assessing the effects of multiple GCMs to the climate response function, the limitation of selecting one to few GCMs, which is a major drawback of typical climate change impact assessments, can be reduced. Descriptions of the stochastic model and basin models are in the following sections. 3.2 Stochastic Generation of Climatic Observations The Markov chain model and the first-order vector autoregressive (VAR1) model for multiple sites formulated by Wilks (1999) are used for the 10,000-year generation of precipitation and temperature. Previous results showed good performance in reproducing mean, variance, and spatial correlation of observations at 62 stations located in Oregon and Idaho, and the methodology has been widely used for various purposes, including the basis of many weather generation studies (e.g., Qian et al. 2002; Brissette et al. 2007; Srikanthan and Pegram 2009; Keller et al. 2015). The precipitation model uses the first-order Markov chain and the mixed exponential distribution for occurrence and amount of precipitation, respectively. To reproduce observed spatial correlations between gauging stations, Wilks (1999) incorporated multivariate Gaussian random numbers into the Markov Chain model. The

28 17 correlation matrix of the multivariate Gaussian distribution is obtained by observed spatial correlations and a generic regression with an exponential function and distances between gauging stations. Besides, the analytical solution of the VAR1 model is used for generating temperatures. The time-series of daily maximum and minimum temperatures are standardized with single-wave cosine functions for each station. Dry and wet days are separately standardized using precipitation generated by the Markov-chain model to stipulate lower averages on wet days. Although a matrix adjustment was necessary for the analytical parameter estimation in the case of 62 stations in the original study of Wilks (1999), the 31 stations of this study did not require the adjustment process. Further details of the multi-site daily weather generation model are found in Wilks (1999). 3.3 Estimation of Annual Water Supply For simulating natural flows from 23 watersheds, we used the semi-distributed tank model proposed by Kim and Kaluarachchi (2014) that includes SNOW-17 of the National Weather Service (Anderson 1976) to simulate the snowmelt process. SNOW-17 has 12 parameters (5 for snowmelt, 7 for runoff) and uses daily precipitation, and maximum and minimum temperatures as inputs. It divides a watershed into five zones with the elevation profile, and thus considers the effects of elevation on precipitation and temperatures. The runoff component (i.e., watershed response to rainfall and snowmelt) of the model is conceptualized by a two-layer tank proposed by Cooper et al. (2007). Kim and Kaluarachchi (2014) calibrated the semi-distributed tank model for 13 gauged

29 watersheds in or adjacent to the Sevier River Basin using a genetic algorithm, and found consistent performance for simulating natural flows across the upper Sevier River Basin. 18 For estimating annual water supply, we used the proximity-based regionalization approach for 23 watersheds in this study. In other words, the calibrated parameter sets of 13 gauged watersheds in Kim and Kaluarachchi (2014) were transferred to the 23 watersheds of this study based on proximity. Hence, the streamflow in each watershed was simulated using the 10,000-year precipitation and temperature values and transferred parameter sets. The simulated streamflows from the 23 watersheds were finally summed up for representing the basin-wide natural water supply. 3.4 Estimation of Irrigation Water Requirements FAO AquaCrop (Steduto et al. 2009) was used to estimate the water requirements for major crops (alfalfa, maize, barley, and wheat) in the eight farming regions. FAO AquaCrop has been frequently used and provided reliable estimates of water requirements for major crops in prior studies (e.g., Araya et al. 2010; Stricevic et al. 2011; Vanuytrecht et al. 2014) based on its simplicity and robustness. It simulates crop yield response to water and salinity on a daily basis, and provides yield estimates per a given irrigation management schedule. Iterative simulations with changing irrigation schedules provide seasonal water requirement and corresponding attainable crop yield. Kim and Kaluarachchi (2015, 2016) calibrated the parameter sets for the target crops in farm lands near Delta using Landsat images and regional crop information. We used these calibrated parameter sets for the eight farming regions of this study.

30 19 For estimating basin-wide agricultural water demand, we first determined the best irrigation schedule by testing 10 irrigation intervals (3-30 days with steps of 3 days) per seasonal irrigation depth. The maximum yields among the 10 schedules is considered to be the yield at the given seasonal irrigation depth. We assumed the crop yield at 1,500 mm of seasonal irrigation depth is the attainable yield, and determined the required irrigation depth for each crop at which the simulated yield becomes 95% of the attainable yield (Figure 3). Figure 3. A graph illustrating the algorithm for calculating the agricultural water requirement in AquaCrop. The agricultural water demand for each crop in each farming region was estimated by multiplying the simulated irrigation requirements and planting areas. Seasonal

31 20 irrigation requirements for pasture and grass hay, which are unavailable in AquaCrop, were estimated using the approach of Masoner et al. (2003), and included as an additional water demand. The basin-wide agricultural water demand is obtained by summing up all demands in the eight farming regions. Because this complex simulation is very timeconsuming when using all 10,000-year climatic inputs in AquaCrop, we simulated agricultural water demand for 100-year inputs first and developed a regression model between agricultural water demand and climatic conditions. The 10,000-year agricultural water demands are estimated by the regression model: D t = P c T cmax (R 2 = 0.75) (2) where Dt is annual agricultural water demand (Mm 3 ), t is time (years), Pc and Tcmax are total precipitation (mm) and maximum temperature ( C) during the growing season (April to September), respectively. 3.5 GCMs and Downscaling Methods The projections of precipitation and temperatures were obtained from GCMs from the coupled model intercomparison project phase 5 (CMIP5). The projections were bias corrected and used statistically downscaled datasets as described by Maurer et al. (2007) for the period Two Representative Concentration Pathways (RCPs), RCP4.5 and RCP8.5, were used to accommodate both intermediate and very high greenhouse gas concentrations, respectively., RCP4.5 and RCP8.5 emission scenarios are commonly used in many climate change assessment studies (e.g., Yan et al. 2015, He et al and

32 Zhang 2016). The labels 4.5 and 8.5 refer to the expected anomaly in radiative forcing in watt/m 2 by year As reported in the IPCC 2013 report, RCP4.5 is characterized by stabilizing the greenhouse gases concentration after year 2100, and the emissions peak occurs around year 2040 and declines thereafter. It assumes a medium population growth, higher rate of technology development, cropland area decrease, and emission mitigation policy to decrease the fossil fuel use. The average global temperature in RCP4.5 is projected to increase by 1.1 to 2.6 C and the global sea level is projected to increase by 0.32 to 0.63 m, relative to the average. While, RCP8.5 is characterized by increasing the green gas concentration over time with continued increase after It assumes a high population growth, lower rate of technology development, cropland area increase, and the global use about 80% of energy from fossil fuels. The average global temperature in RCP4.5 is projected to increase by 2.6 to 4.8 C and the global sea level is projected to increase by 0.45 to 0.82 m, relative to average. In this study, the numbers of GCMs used for RCP4.5 and RCP8.5were 31 and 29 respectively. The climate models used for each scenario and the information for each model are provided in Table 3. These projections were used to project the future basin agricultural water shortage under the climate change.

33 Table 3. Information of 31 and 29 GCMs used in the analysis for RCP4.5 and RCP8.5 scenarios respectively, with NA indicating non-availability. Adapted from Program for Climate Models Diagnosis and Intercomparison (PCMDI). GCM ACCESS1.0 Emission Path RCP4.5 RCP8.5 * * BCC-CSM1.1 * * BCC-CSM1.1(M) * * CANESM2 * * CCSM4 * * CESM1-BGC * * CESM1-CAM5 * * CMCC-CM * * CNRM-CM5 * * CSIRO-MK3.6.0 * * FGOALS-G2 * * FIO-ESM * * Institution CSIRO (Commonwealth Scientific and Industrial Research Organisation, Australia), and BOM (Bureau of Meteorology, Australia) Beijing Climate Center, China Meteorological Administration Beijing Climate Center, China Meteorological Administration Canadian Centre for Climate Modelling and Analysis National Center for Atmospheric Research National Science Foundation, Department of Energy, National Center for Atmospheric Research National Science Foundation, Department of Energy, National Center for Atmospheric Research Centro Euro-Mediterraneo per I Cambiamenti Climatici Centre National de Recherches Meteorologiques / Centre Europeen de Recherche et Formation Avancees en Calcul Scientifique Commonwealth Scientific and Industrial Research Organisation in collaboration with the Queensland Climate Change Centre of Excellence LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences; and CESS, Tsinghua University The First Institute of Oceanography, SOA, China GFDL-CM3 * * Geophysical Fluid Dynamics Laboratory GFDL-ESM2G * * Geophysical Fluid Dynamics Laboratory GFDL-ESM2M * * Geophysical Fluid Dynamics Laboratory 22

34 23 Table 3 (Continued). GCM Emission Path RCP4.5 RCP8.5 Institution GISS-E2-H-CC * NA NASA Goddard Institute for Space Studies GISS-E2-R * * NASA Goddard Institute for Space Studies GISS-E2-R-CC * NA NASA Goddard Institute for Space Studies HADGEM2-AO * * National Institute of Meteorological Research/Korea Meteorological Administration HADGEM2-CC * * Met Office Hadley Centre HADGEM2-ES * * Met Office Hadley Centre and Instituto Nacional de Pesquisas Espaciais INM-CM4 * * Institute for Numerical Mathematics IPSL-CM5A-MR * * Institut Pierre-Simon Laplace IPSL-CM5B-LR * * Institut Pierre-Simon Laplace MIROC-ESM * * Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies MIROC-ESM-CHEM * * Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies Atmosphere and Ocean Research MIROC5 * * Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine- Earth Science and Technology MPI-ESM-LR * * Max Planck Institute for Meteorology (MPI-M) MPI-ESM-MR * * Max Planck Institute for Meteorology (MPI-M) MRI-CGCM3 * * Meteorological Research Institute NORESM1-M * * Norwegian Climate Centre

35 24 CHAPTER 4 4. RESULTS AND DISCUSSION 4.1 Climate Response Function The 10,000-year stochastically generated precipitation and temperature profiles were used to estimate basin-wide annual water availability (St) and agricultural water demand (Dt). A multiple regression model was used to establish the climate response function (i.e. the relationship between It and representative precipitation and temperatures) as shown in Equation (3) with a R 2 of log( I t + 1.8) = P c P o T 2 cmax (3) where, Pc and Tcmax are total precipitation (mm) and maximum temperature ( C) during the growing season (April to September), respectively, and Po is total precipitation (mm) during off-season (October to March). All independent variables were significant (pvalues << 0.001), and the regression model showed no multicollinearity based on its variance inflation factor of less than 2. The climate response function indicates that higher precipitation and lower maximum temperature during a growing season result in less basin-wide agricultural water demand. This is because higher rainfall and lower temperature reduces irrigation water requirement for attainable yields. Off-season precipitation is mostly snowfall that becomes natural water supply in the upcoming growing season, and therefore it is highly sensitive to off-season precipitation. Based on the climate response function, the system is more sensitive to off-season precipitation change than precipitation during the growing

36 25 season. An increase of 5% in Po would result in an increase of It by 46%, while the same change in Pc increases It by only 14%. An increase of 1 C in Tcmax would decrease It by 17%. Overall, Po is the most important parameter affecting basin-wide agricultural water scarcity, and Tcmax and Pc are the next most important parameters. 4.2 Climate Change Impacts The projected precipitation and temperature from GCMs were used to estimate It and statistically analyzed for four periods: , , and with scenarios RCP4.5 and RCP8.5. Each period has 25 values of It. In general, RCP4.5 and RCP8.5 scenarios show an increase in annual precipitation and temperature. Table 4 shows the percent of change of precipitation and temperature for the four time periods: , , and for both scenarios RCP4.5 and RCP8.5. Both scenario projections have an increasing trend in Po and Tcmax. The annual mean temperature rises according to both scenarios by 2.5 C and 4.9 C under RCP4.5 and RCP8.5, respectively at the end of 21st century. Pc has a decreasing trend under RCP8.5 but it does not have a similar consensus under RCP4.5. Figures 4 through 6 show the changes in the three variables of the climate response function: Po, Pc, and Tcmax respectively, under the RCP4.5 scenario. These findings are consistent with the projections of precipitation and temperature estimates from previous studies (e.g., Christensen and Lettenmaier 2006). Christensen and Lettenmaier (2006) found that the mean temperature in Colorado River Basin will increase in the period by 2.7

37 26 and C for less and high emissions scenarios, respectively. They also found that winter precipitation will increase. Table 4. Percent change in future precipitation and temperature for RCP4.5 and RCP8.5 scenarios. dpc and dpo are the percent changes in growing season precipitation and offseason precipitation, respectively. dtcmax and dt are the changes in maximum temperature during the growing season and annual average temperature, respectively. Scenario Parameter dpc (%) RCP4.5 dpo (%) dtcmax ( 0 C) dt( 0 C) dpc (%) RCP8.5 dpo (%) dtcmax ( 0 C) dt ( 0 C) Figure 4. Projections of precipitation during the off-season (Po) for the RCP4.5 scenario.

38 27 Figure 5. Projections of precipitation during the growing season (Pc) for the RCP4.5 scenario. Figure 6. Projections of maximum temperature during the growing season (Tcmax) for the RCP4.5 scenario.

39 28 Figure 7 shows the estimated box plots of It values obtained from the GCMs under RCP4.5 with similar results for RCP8.5 in Figure 8. Although all time periods have GCMs with negative It values for both scenarios, mean values indicate no water scarcity even under changing climate. Indeed, there is an increasing trend in the average of It values, and this is because of increasing Po under both scenarios. Greater It values were obtained under RCP8.5 than RCP4.5 due to higher Po values with RCP8.5. From the results, it is evident that more natural water supply can be expected in the Sevier River Basin due to increasing off-season precipitation, but rising maximum temperature during growing season can increase higher irrigation requirements as well. Figure 7. Box plot of future agricultural water deficit index (It) predicted by GCM projections for RCP4.5.

40 29 Figure 8. Box plot of future agricultural water deficit index (It) predicted by GCM projections for RCP8.5. One advantage of the climate response function is to identify the climatic conditions affecting water availability through values below the threshold (It < 0). Since the climate response function has three predictor variables (see Equation 3), it is not possible to provide a single plot. Instead plots were made with two variable while keep the third one constant. Figure 9 shows the plot with constant off-season precipitation (Po), while Figure 10 shows the same with constant growing season precipitation (Pc). In Figure 9, the points are the average of growing season maximum temperature and precipitation over the period for the 31 and 29 GCMs from the RCP4.5 and RCP8.5 scenarios, respectively. Similarly, the points in Figure 10 are the average of

41 30 growing season maximum temperature and off-season precipitation. GCM projections from both scenarios are imposed with the contour lines of the agriculture water deficit index to identify the GCMs that have unacceptable thresholds (It < 0). There are number of GCM projections in both scenarios that are below the threshold (It < 0), which are to the left to the zero contour line. These climatic conditions are problematic and have the potential of producing water scarcity or climate risk in the different time periods. As shown in Figure 9, there are 1 out of 31 GCMs for RCP4.5 and 2 out of 29 GCMs for RCP 8.5 that produce climate risk. Similarly, in Figure 10, there are 4 out of 31 GCMs for RCP4.5 and 6 out of 29 GCMs for RCP8.5 that produce climate risk. In general, it can be stated that the results indicate lower probability of climate risk for water scarcity with both emission scenarios. After estimating the agricultural water deficit index, the values are compared with the threshold (It=0) to calculate the probability of exceeding the threshold. The cumulative probability of climate conditions was estimated by using a nonparametric empirical distribution presented by Wilks 1995, as shown in Equation 4. The cumulative probability of X variable is the probability that X will take a value less than or equal to x. Each GCM run is assumed to have an equal probability of occurrence in the future. P = m n+1 (4) where P is cumulative probability, m is rank of data, and n is the total number of data values.

42 31 Change in Annual Max Temperature During The Growing Season (C) Change in Annual Mean Precipitation During The Growing Season (%) RCP4.5 RCP Figure 9. Scatterplot of % change in annual mean precipitation and annual maximum temperature during the growing season, superimposed on the contour lines of It for RCP4.5 and RCP8.5 scenarios when Po is constant at mm.